Implement “Happy Eyeballs” algorithim (RFC 8305) in socket.create_connection()

[origin400-p]

BPO	44644

^{Note: these values reflect the state of the issue at the time it was migrated and might not reflect the current state.}

Show more details

GitHub fields:

assignee = None
closed_at = None
created_at = <Date 2021-07-15.12:05:20.908>
labels = ['library', '3.9', 'performance']
title = 'Implement \xe2\x80\x9cHappy Eyeballs\xe2\x80\x9d algorithim (RFC 8503) in socket.create_connection()'
updated_at = <Date 2021-07-15.12:05:20.908>
user = 'https://bugs.python.org/origin400-p'

bugs.python.org fields:

activity = <Date 2021-07-15.12:05:20.908>
actor = 'origin400-p'
assignee = 'none'
closed = False
closed_date = None
closer = None
components = ['Library (Lib)']
creation = <Date 2021-07-15.12:05:20.908>
creator = 'origin400-p'
dependencies = []
files = []
hgrepos = []
issue_num = 44644
keywords = []
message_count = 1.0
messages = ['397542']
nosy_count = 1.0
nosy_names = ['origin400-p']
pr_nums = []
priority = 'normal'
resolution = None
stage = None
status = 'open'
superseder = None
type = 'performance'
url = 'https://bugs.python.org/issue44644'
versions = ['Python 3.9']

[origin400-p]

While support for the so-called “Happy Eyeballs” algorithim described in RFC 8305 was implemented for asyncio in Issue bpo-33530, socket's create_connection function remains left without it causing suboptimal performance in broken dual-stack environments.

[gpshead]

socket.create_connection() does indeed need this feature. In addition to our existing asyncio implementation's behavior we could also look at what Golang has done, they implemented the original rfc-6555 quite a while back and are tracking 8305 in golang/go#23841. (even only implementing rfc-6555's variant would be a big improvement)

[gpshead]

It'd also be interesting if we could avoid maintaining two implementations of this in the stdlib so long as the cost to do so is not even more complicated to maintain. (ex: Would the ironic thing of socket.create_connection running its own singular asyncio event loop just for the duration of this task feasible or a Bad Idea™️ ?)

[graingert]

@gpshead this would then break users that use create_connection inside an existing asyncio event loop

[graingert]

@sethmlarson has a sync implementation of happy eyeballs, it might be an idea to land that in cpython